Extensions

Extensions allow you to configure how your assistant behaves in certain circumstances. Each extension can be activated or deactivated as needed and its various options edited to achieve the specific performance required for your project.

You can access the Manage Extensions page from either the More menu (to the right of each project) on the Project list page (to the right of each project) or from the More menu (next to the project's title) within a project itself. All available extensions, and those that are coming soon, will be shown. Select one of the following cards to enable/disable and configure the extension :-

  • Short Tail - Used to deliver a single response from the when your assistant is confident it has the right answer

  • Long Tail - Used to offer a list of multiple responses when your is not confident it has a single right answer

  • Feedback - Used to ask the end-user for feedback on the usefulness of the last response

  • Spelling - Used to correct spelling errors and configure a custom dictionary for the project

Short Tail

Short Tail refers to the delivery of a (usually single) response to the end-user's utterance when your assistant's Natural Language Classifier has matched it to an intent in the knowledge base with a high enough confidence, so answering the question or triggering the required action.

Occasionally the classifier may match more than one intent with a high enough confidence i.e. finds multiple responses it deems an appropriate answer. When this occurs, the end-user will be presented with a list of intents that have been confidently matched. They will be able to choose which one they think is the most appropriate and then see that intent's response.
Multiple short tail responses are a sign that multiple intents are conflicting. This usually requires either deletion of duplicate intents or splitting one intent with overly general content into multiple intents with more focus.

If no intents are matched with a high enough confidence then a Low Confidence response is given, usually informing the end-user that the assistant either does not understand the question or does not know the answer.

You can choose to toggle this extension on/off here.

Confidence Levels

You can change the confidence levels that dictate how intents are matched by the classifer -

  • Minimum Confidence
    The minimum confidence score where an intent will be matched to an end-user's utterance and therefore returned. This is set to 80 as default, but can be changed to best suit your project.

  • Elegant Fail Confidence
    Any intents matched at this confidence score or below will not be returned to the end-user. This is set to 25 as default, but can be changed to best suit your project.

    Any intents that fall in the middle, i.e. are matched at less than the Minimum Confidence and more than the Elegant Fail Confidence scores, may then trigger a Long Tail search. Read more about the Long Tail extension below.

Custom Replies

Some short tail results involve explanatory responses. Each type of result comes with a default response, which can be edited if necessary.
New Custom Replies can also be added to each type of result. If more than one exists per response type, these are returned randomly.

  • Low Confidence
    Displayed when the end-user's input cannot be accurately matched to any intents within the knowledge base e.g. I'm sorry, I don't understand what you're asking...

  • Multiple Short Tail Introduction
    Used to introduce the list of multiple intents that have been determined by the classifier to be high-confidence matches e.g. I'm confident all of these answers can help to answer your question:

Click Save Extension to save any changes you've made.

Long Tail

Long Tail allows a conversation to continue even when a end-user's utterance cannot be matched to an answer or action with a high confidence. If the Short Tail classifier cannot return a high confidence match then a Long Tail search can deliver a list of intents that may contain information that will help. Long Tail uses not only an intent's User Intents as training material, but also the contents of its Response.
A Long Tail search can also be triggered by an end-user providing negative feedback to a short tail response, in an attempt to continue the conversation.

You can choose to toggle this extension on/off here.

Custom Replies

Some long tail results involve explanatory responses. Each type of result comes with a default response, which can be edited if necessary. New Custom Replies can also be added to each type of result. If more than one exists per response type, these are returned randomly.

  • Long Tail Introduction
    Displayed to introduce the list of intents returned by the Long-tail search e.g. I'm not confident I can answer that yet. Here's some questions I do understand that I think may be related.

  • Direct Long Tail Introduction
    Displayed to introduce the intent returned by the Long-tail search (if only one was found) e.g. I believe this is what you're looking for:

  • Short Tail to Long Tail Crossover
    Displayed to introduce the list of intents returned by the Long-tail search that was triggered when a short tail response was met with negative feedback e.g. Sorry, I didn't understand the question fully. Try one of these:

  • Long Tail Fail
    Displayed when an end-user has indicated that the Long-tail results provided are incorrect (by giving negative feedback) e.g. Sorry, about that I thought I knew the answer...

Click Save Extension to save any changes you've made.

Feedback

The Feedback extension allows you to prompt the end-user for positive or negative feedback to the responses they receive from your assistant.

You can choose to toggle this extension on/off here.

Custom Replies

There is only one type of custom reply for feedback. There is a default response, which can be edited if necessary. New Custom Replies can also be added. If more than one reply exists, these are returned randomly.

  • Feedback
    Displayed to introduce the feedback options that follow a response from the assistant e.g. Was that answer helpful?

Click Save Extension to save any changes you've made.

Spelling

The Spelling extension allows you to turn on automatic spelling correction of end-users' inputs. You can manage a custom dictionary to add and edit spelling exclusions and spelling error replacements.

You can choose to toggle this extension on/off here.

Dictionary

To start customising your spelling dictionary, click Create Term

Spelling Exclusion

Add a term that you want to ensure remains unaffected by automated spelling corrections.

Enter in the sidebar -

  • Enter Term
    The term to be reserved from spellchecking e.g. ABC Corp

  • Substitute Terms
    Leave this field blank

    Now, the term ABC Corp in an end-user's utterance will be protected from any spelling corrections enforced by the Spelling extension.

Spelling Error Replacement

Add one or more terms that you want to be automatically changed to another term.

Enter in the sidebar -

  • Enter Term
    The base term that the substitute terms will be corrected to e.g. People Team

  • Substitute Terms
    A term that is to be changed to the base term e.g. Human Resources

    Continue adding more substitute terms as needed e.g. HR

    Now, the terms Human Resources and HR in end-user utterances will be automatically corrected to People Team.

Click Save Extension to save any changes you've made.

Still need help? Get in touch!
Last updated on 5th Jun 2020